Automate entire workflows with AI, not just pieces
I build production AI automation that replaces hours of manual work — multilingual email processing, document extraction, lead research, content generation, multi-step agent pipelines. Delivered on Make.com, n8n, or custom code, whichever fits the job.
- Typical build: 2–4 weeks from proposal to handover
- Works with your existing tools — no rip-and-replace
- You own all accounts, all data, all configs
Who this is for
If a chunk of your team's time goes to repetitive work that AI could handle — reading emails, pulling data from PDFs, researching leads, triaging tickets, generating reports — this is the service.
Operations leaders
Your team is hitting a ceiling because growth means more manual work. You need automation that actually ships, not another Zapier graveyard.
Travel & hospitality operators
Manual hotel sourcing, supplier email chains, multilingual inquiries. I build AI-driven sourcing and comparison systems (like the one built for Navitours).
Agencies & B2B services
Lead research, proposal drafting, client reporting. High-leverage AI automation that lets you take on 2x the clients without 2x the staff.
SMB founders
You don't need enterprise software, you need someone to wire up AI into your existing workflow and make it boringly reliable.
Why most AI automation fails in production
The gap between "ChatGPT can do this" and "my team uses this daily" is massive. Most teams I meet have tried to bolt AI onto their workflow with a mix of Zapier, ChatGPT tabs, and shared Google Sheets — and given up after three weeks because it was more work than the manual process.
The reasons are predictable: AI outputs aren't validated, so one bad response breaks the downstream process. Prompts aren't versioned, so when they "stop working" nobody knows why. There's no error handling, so one failed email blocks the whole queue. The automation depends on six tools that don't integrate cleanly. And nobody owns it, so when it breaks on Friday at 5pm, it stays broken.
Real automation isn't about connecting tools — it's about building a reliable pipeline that handles errors gracefully, validates AI outputs, escalates edge cases to humans, logs everything, and runs without supervision for months.
How I build AI automation that runs for years
Every pipeline I ship is designed to run 24/7 without attention — or to fail loudly and clearly when it can't.
Right tool for the job
Make.com for orchestration when speed of iteration matters. n8n when you need self-hosting or more logic. Custom code when neither fits. I don't force everything through one platform — the right tool saves weeks.
AI output validation
Every LLM response is validated against a schema before it moves downstream. If the AI returns incomplete data, the pipeline flags it as "to-verify" rather than failing silently. Nothing makes its way to the customer-facing stage without passing checks.
Human-in-the-loop by design
Edge cases get escalated to a human via Slack, email, or a dashboard — with enough context to decide in seconds. You stay in control of the 5% that matters while AI handles the 95% that doesn't.
Full observability
Every step of every run is logged: inputs, AI outputs, API responses, errors, latency. When something goes wrong three weeks later, you can trace it in minutes. When performance drifts, you see it.
Cost-aware by design
I pick the cheapest model that handles the job. Routing logic sends simple tasks to mini models, complex ones to frontier models. Your OpenAI bill stays predictable.
Your accounts, your data
Every service account (Make.com, OpenAI, Apify, databases) is created in your name. You have full visibility into costs, full access to stop or modify the system, and full ownership of the data flowing through it.
How we work together
Fast, scoped engagements — most automation projects ship in 2–4 weeks.
Discovery
30-min call to understand the manual process, the systems involved, and the ROI target. I ask the questions that uncover hidden edge cases before they become production bugs.
Proposal
Written proposal with architecture, tool choices, cost estimate (build + monthly running), timeline, and fixed price. You see the exact flow before committing.
Build
Core pipeline delivered in week 1. AI extraction, validation, and output formatting in week 2. Live testing across real data in week 3. Handover + documentation in week 4 (if needed).
Handover
Full documentation in Notion or your preferred tool. Loom walkthrough video of how to operate, extend, and debug the system. 14 days of post-launch support included.
Ongoing (optional)
Retainer arrangement if you want continued improvements, new automations added, or monitoring managed. Or self-serve with the documentation — your choice.
What you get
- Complete automation workflow (Make.com scenario / n8n workflow / custom code)
- Service accounts (SerpApi, OpenAI, Apify, etc.) set up in your name
- Google Sheet / database / dashboard for outputs
- AI prompt templates with version history
- Validation schemas and error handling
- Observability dashboard (run history, costs, errors)
- Full documentation + Loom walkthrough
- 14 days of post-launch support
Investment
Fixed-price based on scope. Most automation projects fall in the ranges below.
Single Workflow
$1,500 – $3,500
One focused automation — email triage, document extraction, lead research, etc. Typical build: 1–2 weeks.
- One complete automation pipeline
- AI extraction with validation
- Error handling + notifications
- Documentation
- 14-day post-launch support
Multi-step System
$3,500 – $8,500
Multiple connected workflows — e.g. the Navitours hotel sourcing system: discovery → email collection → AI inquiry → reply monitoring → comparison sheet. Typical build: 2–4 weeks.
- Multiple connected workflows
- Multilingual AI processing
- Auto follow-up logic
- Comparison / ranking layer
- Custom dashboard or sheet
- Full observability
- Documentation + Loom walkthrough
Custom Build
$8,500+
Complex automation with custom code, integrations with internal systems, or high-volume production requirements.
- Everything in Multi-step
- Custom code for edge cases
- Internal system integrations
- High-throughput architecture
- Production monitoring
- Extended support
Monthly running costs (SerpApi, Make.com, OpenAI, etc.) are separate — typically $50–250/month depending on volume. All accounts are in your name.
Tech Stack
Working With Clients Across
Frequently asked
How is this different from Zapier or a no-code tool?
Zapier is fine for trigger-action chains. Real AI automation needs validation, branching, human-in-the-loop, error recovery, and cost control — things I build properly in Make.com, n8n, or custom code. The difference is whether it breaks after three weeks or runs for years.
What if the AI gives wrong answers?
Every AI output is validated against a schema before it moves forward. Outputs that don't pass get flagged for human review, not silently passed through. Plus, I run accuracy tests across real data before handover, so you see the pass rate before go-live.
Do you work with Greek/Italian/Arabic/other non-English data?
Yes. Multilingual AI extraction is something I specialize in — the Navitours system handles Greek, Italian, and English hotel replies. OpenAI and Anthropic both handle non-English content well, and I tune prompts per language.
Can you integrate with our CRM / internal systems?
Yes. If there's a REST API, I can integrate it. For CRMs like HubSpot, Salesforce, Pipedrive, Attio — most have mature APIs. For custom internal systems, I'll build the integration layer.
What happens if the AI provider has an outage?
Pipelines include retry logic and fallback models. If OpenAI is down, the system can fall back to Anthropic (or vice versa) for critical paths. Non-critical paths pause and resume when service recovers.
Who pays for the API usage — you or us?
You. All service accounts (OpenAI, Make.com, SerpApi, etc.) are set up in your name. You have full visibility into costs and can stop or change anything without needing me. My pricing is build-only; monthly running costs are separate and transparent.
Ready to start?
Every engagement starts with a 30-minute discovery call. I'll listen, ask sharp questions, and send a written proposal within 48 hours.